Tokenizing strings on digit/word boundaries

[polyfloyd]

Hi!

I am working on integrating Meilisearch in our product and have come to the conclusion that our users perform a lot of queries for numeric terms that are not surrounded by separator tokens, but whole words.

Example:

• 186941 should find 123XYZ186941
• 2110063 should find p2110063

From my understanding of Meilisearch internals, these queries do not return these results because the search term does not occur at the start of the tokens to be matched.

The solution I would propose is to have digit/word/? boundaries be counted as token separators. So e.g. 123XYZ186941 would be split into 123, XYZ, 186941. The last token in this series would match the search query.

We are currently working around this limitation by inserting known separators in strings before sending them off to Meili for indexing, but this has as disadvantage that the returned highlighting information no longer matches the original text.

[macraig]

Hi @polyfloyd, we have an experimental CONTAINS filter operator that would return the results you mention.

Would that solve your use case or do you need Meilisearch to somehow separate letters from numbers without a separator?